Get security guidance for Generative AI, LLMs, agents, MCP, and related technologies grounded in the published research of the OWASP GenAI Security Project.
The GenAI Security Advisor is an Agent Skill that gives your AI coding assistant access to a curated, versioned knowledge base of OWASP GenAI Security Project research rather than relying only on the model's general knowledge.
It can help you research, assess, explain, and apply GenAI security guidance across areas including:
- LLM security
- Agentic AI security
- Data security for GenAI
- MCP security
- AI red teaming
- GenAI governance
- Security frameworks and mappings
- Secure AI adoption
- Related OWASP GenAI Security Project research and guidance
In short: install the skill once, then ask your AI assistant GenAI security questions in your normal workflow.
Once installed, you can ask your AI assistant questions such as:
What are the primary security risks associated with prompt injection?
Explain the security risks of excessive agency in an AI agent.
What security concerns should I consider when using MCP?
Review this GenAI architecture and identify the security risks I should consider.
What GenAI security controls should I consider for an application
that uses an LLM with access to internal company data?
How does this application map to the OWASP LLM Top 10?
Which OWASP GenAI security guidance applies to this AI agent?
What security controls should I implement to mitigate this risk?
Compare the security risks of an LLM application and an autonomous AI agent.
What are the recommended mitigations for this GenAI security issue?
You can provide code, architecture diagrams, configuration, requirements, or other project information and ask the Advisor to help identify relevant GenAI security considerations.
For example:
Review this code for GenAI security risks and identify which
OWASP GenAI guidance is relevant.
Here is the architecture for our AI agent. Identify potential
security risks and recommend mitigations.
GenAI security guidance changes quickly. General-purpose AI models may have incomplete, outdated, or inconsistent knowledge of the latest OWASP GenAI Security Project research.
The Advisor addresses this by grounding its guidance in a versioned, curated corpus of OWASP GenAI Security Project research.
The corpus covers areas such as the LLM Top 10, Agentic Top 10, Data Security risk and framework mappings, MCP security, red teaming, and governance guidance.
The corpus is intentionally maintained as point-in-time copies rather than live links to upstream repositories. This means the guidance you receive does not unexpectedly change because an unrelated upstream document was modified.
The GenAI Security Advisor works with multiple Agent Skill-compatible tools.
| AI Assistant | Supported |
|---|---|
| Claude Code | Yes |
| OpenAI Codex CLI | Yes |
| GitHub Copilot | Yes |
| Gemini CLI | Yes |
The skill uses the open Agent Skills format, with a single canonical SKILL.md that can be discovered by supported platforms.
To install the skill use the method for your AI assistant.
Use the repository's plugin marketplace:
/plugin marketplace add GenAI-Security-Project/GenAI-Security-Advisor
Then install the skill:
/plugin install
Alternatively, copy the skill into your Claude skills directory:
~/.claude/skills/genai-security-advisor/
Install the skill globally:
npx skills add GenAI-Security-Project/GenAI-Security-Advisor -gThe same installer can target multiple supported hosts, for example:
npx skills add GenAI-Security-Project/GenAI-Security-Advisor \
-g \
-a codex \
-a claude-code \
-a copilot \
-a gemini-cliInstall the Gemini extension:
gemini extensions install https://github.com/GenAI-Security-Project/GenAI-Security-AdvisorInstall the skill with the GitHub CLI:
gh skill install GenAI-Security-Project/GenAI-Security-AdvisorThe skill is installed into:
~/.copilot/skills/
The repository documents these platform-specific registration methods. Because AI coding tools and their CLI interfaces can change, check the current tool documentation or --help output if an installation command has changed.
Once the skill is installed, you don't need to learn a new command language.
Just ask your AI assistant a GenAI security question.
For example:
What OWASP GenAI security guidance should I consider when
building an AI agent that can execute tools?
Or:
Review this AI agent design and identify the security risks.
For each risk, explain the relevant OWASP guidance and
recommended mitigations.
Or:
I'm building an MCP-enabled application. What security
controls should I consider?
The skill determines which research resources are relevant and uses the included corpus to ground its response.
The more context you provide, the more useful the Advisor can be.
Instead of:
How do I secure AI?
Try:
I'm building an internal enterprise application that uses an LLM
to summarize confidential documents. The model can retrieve
documents from our internal knowledge base.
What are the key GenAI security risks I should consider?
Map the risks to relevant OWASP GenAI guidance and recommend
practical mitigations.
Useful context can include:
- Your application architecture
- The type of AI model you're using
- Whether you're using agents
- Tools available to the model or agent
- Data sources
- Authentication and authorization mechanisms
- MCP servers
- External APIs
- Deployment environment
- Trust boundaries
- Security requirements
- Code or configuration
When security guidance matters, you can ask the Advisor to explain where its recommendation comes from.
For example:
Which OWASP GenAI Security Project resource supports this recommendation?
Show me the relevant OWASP GenAI security guidance for each risk.
Explain how this recommendation maps to the OWASP LLM Top 10.
This can help you distinguish OWASP GenAI Security Project guidance from general security recommendations.
The repository contains the skill itself and a curated research corpus.
This is the Agent Skill's instruction set. It tells your AI assistant how to use the GenAI Security Advisor and how to work with the included research.
There is one canonical SKILL.md at the repository root. Platform-specific paths point back to this canonical version.
This contains the research resources used by the Advisor.
The resources are maintained as frozen, point-in-time copies so that the Advisor's knowledge does not unexpectedly change when an upstream repository changes.
The corpus/MANIFEST.yaml file provides information about the resources included in the corpus, including their status and licensing.
The research corpus is intentionally versioned.
Updates to upstream OWASP GenAI Security Project resources are not automatically pulled into your installation. Updates are reviewed and incorporated into the Advisor's corpus deliberately.
This design provides a more predictable experience: the guidance you're using is tied to a known version of the Advisor rather than silently changing underneath you.
When a new version of the Advisor is released, update your installed skill using the update mechanism provided by your AI assistant.
The GenAI Security Advisor is intended to help you understand and apply GenAI security research and guidance.
It should not be treated as a substitute for:
- Your organization's security policies
- Security architecture review
- Threat modeling
- Penetration testing
- Legal or regulatory review
- Risk acceptance processes
- Professional security expertise
Use the Advisor as a research and decision-support tool, and validate recommendations against your specific environment and requirements.
The GenAI Security Advisor is an open project and welcomes contributions from the community.
If you have questions about using the GenAI Security Advisor, installing the skill, interpreting its guidance, or getting started, join the OWASP GenAI Security Project Slack and use the:
#team-genai-security-advisor channel
This is the recommended place to ask questions, discuss how the Advisor works, and connect with others using or contributing to the GenAI Advisor project.
Contributions are welcome, including:
- Improving the Agent Skill instructions
- Adding or improving examples
- Identifying gaps or issues in the research corpus
- Suggesting additional OWASP GenAI Security Project resources
- Improving documentation
- Reporting bugs or unexpected behavior
- Proposing enhancements to the Advisor
Before making a contribution, consider discussing the proposed change in #team-genai-security-advisor so the community can provide feedback and help coordinate the work.
When contributing research or other material to the corpus, please review the applicable licensing information in corpus/MANIFEST.yaml and preserve the original licensing requirements.
The GenAI Security Advisor repository contains both project-owned content and third-party research content.
The repository's own content—including SKILL.md, scripts, workflow files, manifest files, and this README—is licensed under Apache-2.0.
Content in corpus/ retains its original license. Much of the included OWASP GenAI Security Project content is licensed under CC BY-SA 4.0.
If you redistribute material from the corpus, check corpus/MANIFEST.yaml for the applicable license for each resource.
The current public release is v1.1.0.
The project follows Semantic Versioning. See CHANGELOG.md for release history.
If you just want to get started:
- Install the GenAI Security Advisor for your AI assistant.
- Open your project in that AI assistant.
- Ask a GenAI security question.
- Provide your architecture, code, or other context when you want a more specific assessment.
- Ask for the relevant OWASP guidance and mitigations when you need to understand the basis for a recommendation.
That's it.
The GenAI Security Advisor brings OWASP GenAI Security Project research into the AI-assisted development workflow you're already using.